Add warning about LS-based optimizers with the EM method.

Project.toml

@@ -1,7 +1,7 @@
 name = "Vecchia"
 uuid = "8d73829f-f4b0-474a-9580-cecc8e084068"
 authors = ["Chris Geoga <[email protected]>"]
-version = "0.7.0"
+version = "0.7.1"
 
 [deps]
 BangBang = "198e06fe-97b7-11e9-32a5-e1d131e6ad66"

src/em_iterator.jl

@@ -86,6 +86,7 @@ function vecchia_mle_withnugget(cfg, init, optimizer; optimizer_kwargs...)
 end
 
 function em_estimate(cfg, saa, init; 
+                     warn_optimizer=true,
                      verbose=true,
                      optimizer=ipopt_optimize, 
                      norm2tol=1e-2,
@@ -108,6 +109,14 @@ function em_estimate(cfg, saa, init;
       Maybe you should check your model/code/data basic things again before continuing."
     end
   end
+  if optimizer==ipopt_optimize && warn_optimizer
+    @warn "Ipopt is the default optimizer here, but please note that for large \
+    problems a trust-region based method, such as KNITRO, can work much better. \
+    Since that software is neither gratis or libre Ipopt is the default, but if \
+    you are not happy with the performance you might even consider something like \
+    Optim.jl's NewtonTrustRegion solver or something. Feel free to open an issue \
+    to discuss more."
+  end
   # now use the iterator interface:
   (init_result=mle_withnugget,
    em_refine(cfg, saa, mle_withnugget.minimizer; optimizer=optimizer,